Neuron Explorer

Qwen3-0.6B

'and' joining two words

A neuron in layer 3 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron lights up on the word 'and' when it links two things of the same kind, like 'women and girls' or 'legal and regulatory'. When it was turned up, the model fell into repeating phrases such as 'and that, and that'.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    2 million people a year, most of them women and children. Thisremains a major health concern and equitable human development cannot be achieved as long as women and girls continue to lack access to modern energy sources. Clean and efficient cooking energy has

  2. #2 peak

    women who lost weight. Those hormone levels dropped 16.2 percent in women who dieted and 20.3 percent in women who dieted and exercised, according to the study.⏎"The biggest effect was through diet

  3. #3 peak

    this report, the different legal and regulatory frameworks for the travel and tourism sector are considered which affect the business. Surface, sea and air transport laws within the legal and regulatory

  4. #4 peak

    not read nor write but spoke Irish and English. Margaret could read, write and spoke both Irish and English and Julia could not read nor write but both spoke Irish and English. All were born in County Galway.⏎There were three new columns in

  5. #5 peak

    and college faculty. Many career and technical education programs include opportunities for you to continue your learning in the community through service or employment.⏎To find out more about career and technical education programs of study, talk to your counselor or career and technical education teacher

  6. #6 peak

    History proves that dictatorships do not grow out of strong and successful governments, but out of weak and helpless ones. If by democratic methods people get a government strong enough to protect them

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: it is not, and it is not. It is not, and it is not. It is not, and it is not, and it is not, and it is not, and it is not, and it is not, and

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    they are not invited. The master notes that the people of Dis city have “eternal fire that burns inside them…” (Inferno 8, line, 73). This is meant to warn the narrator that the city dwellers

  2. #2 peak

    7, 72, 76, 77, 81, 86. Our estimates predicts 32 members of this sequence starting from 6.

  3. #3 peak

    cost you to fulfill those orders?⏎If it cost $15,000 to print, store, and ship them to your customers, another $5,000 to maintain your website, $3,000 in miscellaneous

  4. #4 peak

    1918-College of Dentistry organized in Dallas by taking over the State Dental College which had been founded in 1905.⏎1971-The College was separately incorporated in 1971, although Graduate programs

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L0:313 excites "and" joining paired nouns
  2. L1:1250 weakly suppresses
  3. L2:193 weakly suppresses
  4. L2:334 weakly suppresses
  5. L1:17 weakly suppresses
  6. L1:1710 weakly suppresses
  7. L0:2719 weakly suppresses
  8. L1:8 weakly excites
  9. L2:184 weakly excites
  10. L1:2816 weakly suppresses
  11. L0:2770 weakly excites
  12. L1:2867 weakly excites
  13. L0:558 weakly suppresses inside all-caps abbreviations
  14. L1:359 weakly excites
  15. L1:298 weakly suppresses
  16. L1:1583 weakly excites
  17. L2:356 weakly excites
  18. L1:3037 weakly suppresses
  19. L2:86 weakly excites
  20. L1:1122 weakly suppresses
  21. L0:519 weakly excites
  22. L1:923 weakly suppresses
  23. L1:98 weakly excites
  24. L0:2452 weakly excites

Feeds later layers

  1. L4:155 strongly excites
  2. L5:44 strongly excites
  3. L4:559 strongly suppresses
  4. L10:350 strongly excites
  5. L5:119 strongly excites
  6. L4:497 excites
  7. L6:879 suppresses
  8. L6:169 excites
  9. L6:212 excites
  10. L4:483 excites
  11. L7:85 suppresses
  12. L5:29 excites
  13. L7:222 excites
  14. L6:277 suppresses
  15. L6:571 suppresses
  16. L6:258 excites
  17. L10:35 excites
  18. L8:82 suppresses
  19. L5:42 excites
  20. L5:88 excites
  21. L8:73 excites
  22. L7:27 excites
  23. L6:235 excites
  24. L5:107 excites

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

2 values between -0.92 and -0.861 values between -0.86 and -0.803 values between -0.74 and -0.691 values between -0.69 and -0.632 values between -0.63 and -0.573 values between -0.57 and -0.522 values between -0.52 and -0.4611 values between -0.46 and -0.4021 values between -0.40 and -0.3428 values between -0.34 and -0.2965 values between -0.29 and -0.23160 values between -0.23 and -0.17279 values between -0.17 and -0.12530 values between -0.12 and -0.06883 values between -0.06 and -0.001032 values between -0.00 and 0.06496 values between 0.06 and 0.11213 values between 0.11 and 0.17102 values between 0.17 and 0.2340 values between 0.23 and 0.288 values between 0.28 and 0.349 values between 0.34 and 0.406 values between 0.40 and 0.451 values between 0.45 and 0.513 values between 0.57 and 0.632 values between 0.68 and 0.741 values between 0.80 and 0.851 values between 0.97 and 1.031 values between 1.03 and 1.081 values between 1.25 and 1.311 values between 1.31 and 1.37offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. 他们
  2. 器材
  3. 的那个
  4. 的日
  5. 另外一个
  6. 筏
  7. 他们是
  8. ឡ
  9. FLT
  10. lickr

Suppresses

  1. Enough
  2. ends
  3. pering
  4. روس
  5. Vtbl
  6. endez
  7. vis
  8. zeitig
  9. ended
  10. urgeon